How to Write the Methods Section for a Meta-Analysis
By Dr. Zubair Khalid, DVM, MS, PhD ·
Quick Answer
- Use the PRISMA 2020 statement as your structural backbone for describing search, screening, and data extraction steps in the methods section.
- Select and justify your effect size metric, heterogeneity estimators, and sensitivity analyses before writing, then document each choice with a rationale tied to your research question.
- A methods section cannot guarantee reproducibility if you omit software versions, search dates, or eligibility criteria, so treat those details as mandatory instead of optional.
Why the Methods Section Determines Meta-Analysis Credibility
A meta-analysis combines results from multiple independent studies to produce a pooled estimate with greater statistical power than any single study. The methods section is the only place where readers can judge whether that pooling was legitimate. If the methods are vague, the entire analysis becomes unverifiable regardless of how sophisticated the statistics appear.
Peer reviewers and journal editors evaluate meta-analyses with the same scrutiny applied to primary research. They check whether the search strategy could be replicated, whether inclusion criteria were applied consistently, and whether the statistical approach matched the data structure. The EQUATOR Network maintains a library of reporting guidelines that journals increasingly require authors to follow. For meta-analyses of randomized trials, PRISMA 2020 is the standard. For other study designs, alternatives such as MOOSE for epidemiological studies or PRISMA extensions for specific contexts may apply.
The methods section also protects against bias. A reader cannot assess whether you cherry-picked studies or outcomes unless you describe your search dates, databases, and screening process. The Committee on Publication Ethics core practices emphasize that research must be reported honestly and completely. A methods section that omits key decisions makes it impossible to distinguish rigorous work from selective reporting.
Your methods section serves three distinct audiences. The first is the peer reviewer who must verify that your approach was sound. The second is the reader who wants to know whether your conclusions apply to their population or setting. The third is the future researcher who may want to update your analysis or apply your methods to a new question. All three need the same level of detail.
Core Principles of a Reproducible Methods Section
Transparency Over Brevity
Journal word limits often pressure authors to compress methods sections. Resist that pressure. A meta-analysis methods section that is too short to replicate is a failed methods section. The National Library of Medicine hosts research methods texts that consistently emphasize the need for complete procedural descriptions. Every step from question formulation to sensitivity analysis should be traceable.
Structure That Mirrors Your Workflow
The methods section should follow the chronological sequence of your analysis. Start with the research question and eligibility criteria, then describe the search, then screening and data extraction, then risk of bias assessment, then statistical methods. This structure helps readers follow your logic and makes it easier for reviewers to check each step.
Justification for Every Choice
Do not simply state what you did. State why you did it. If you chose a random-effects model, explain that you expected clinical or methodological heterogeneity based on your prior knowledge of the field. If you used a particular heterogeneity estimator, explain why it was appropriate for your number of studies and expected effect sizes.
Distinguishing Protocol From Execution
If you registered your protocol in PROSPERO or another registry, state that clearly and provide the registration number. If you deviated from the protocol, describe the deviations and the reasons. This transparency is central to the Committee on Publication Ethics expectations about honest reporting.
At a Glance
| Methods Component | What to Report | Common Omission |
|---|---|---|
| Search strategy | Full Boolean strings, databases, dates, language limits, hand-searching | Reporting only database names without the actual search strings |
| Eligibility criteria | PICOS elements, inclusion and exclusion rules, study design filters | Describing criteria vaguely without operational definitions |
| Data extraction | Variables collected, extraction forms, dual extraction process, disagreement resolution | Failing to state whether extraction was done independently |
| Risk of bias | Tool used, domains assessed, how ratings informed analysis | Reporting ratings without explaining their analytic role |
| Statistical model | Fixed vs random effects, heterogeneity estimators, effect size metric | Justifying model choice without reference to heterogeneity expectations |
| Sensitivity analyses | Predefined scenarios, rationale for each, how results changed | Running post hoc analyses without declaring them as such |
Preparing Before You Write
Define Your Research Question With PICOS
Every meta-analysis begins with a structured question. The PICOS framework covers Population, Intervention, Comparison, Outcomes, and Study design. Write out each element before you design your search. This step determines your eligibility criteria, your search terms, and your data extraction variables.
For example, if your question concerns the effect of a feed additive on weight gain in broiler chickens, you must specify the bird age range, the additive dose, the comparator (standard feed or a different additive), the outcome measurement timing, and the acceptable study designs. Each of these decisions shapes your methods section.
Register Your Protocol
Protocol registration is now standard practice for systematic reviews and meta-analyses. Registration creates a public record of your planned methods, which makes it harder to change your analysis based on what you find. The National Institutes of Health expects grant applicants to follow rigorous research practices, and protocol registration is part of that expectation for synthesis research.
Your protocol should include your planned search strategy, eligibility criteria, data extraction approach, risk of bias assessment, and statistical analysis plan. When you write your methods section, you will describe what you actually did and note any deviations from the protocol.
Select Your Reporting Guideline
The EQUATOR Network provides a searchable database of reporting guidelines. For a meta-analysis of randomized controlled trials, PRISMA 2020 is the appropriate choice. For other designs, check whether a specific guideline exists. Some journals require a completed checklist submitted with your manuscript, so identify the relevant guideline early.
Writing the Search Strategy Section
Databases and Sources
List every database you searched, the platform used, and the search dates. Common databases for life science research include PubMed, Web of Science, Scopus, and Embase. For animal science topics, you may also need to search Agricola or CAB Abstracts. State whether you searched grey literature sources such as conference proceedings, thesis repositories, or organizational websites.
The National Library of Medicine provides guidance on searching biomedical databases effectively. Your methods section should allow a reader to reproduce your exact search in each database.
Full Search Strings
Include the complete Boolean search string for at least one major database, typically in an appendix or supplementary file. For other databases, describe how the search was adapted. State any filters applied, such as language restrictions, publication date limits, or study design filters.
Additional Search Methods
Describe hand-searching of reference lists, citation tracking, and contact with experts or manufacturers. These supplementary methods can identify studies missed by database searches. State who performed these searches and when.
Search Documentation
Record the number of records retrieved from each source. This information feeds into the PRISMA flow diagram, which shows the flow of records from identification to inclusion. Your methods section should describe how the flow diagram was populated.
Describing Eligibility Criteria
Inclusion Criteria
State your inclusion criteria with enough specificity that a reader could apply them to a study and reach the same decision. Include the population characteristics, intervention details, comparator definitions, outcome measures, and acceptable study designs. If you limited inclusion by language, publication status, or date, state those limits and justify them.
Exclusion Criteria
List the reasons for excluding studies that might otherwise appear relevant. Common exclusions include duplicate publications, insufficient data for effect size calculation, and studies that report outcomes in a format that cannot be converted to your chosen effect size metric.
Operational Definitions
Define terms that could be interpreted differently. For example, if you include studies with a minimum follow-up period, state the exact duration. If you require a specific outcome definition, describe it precisely. These operational definitions prevent ambiguity during screening.
Data Extraction and Management
Extraction Process
Describe who extracted data and whether extraction was performed independently by two reviewers. State how disagreements were resolved, such as by discussion or adjudication by a third reviewer. The Committee on Publication Ethics expects that research records are accurate and verifiable, and dual extraction is a key quality control measure.
Variables Collected
List the variables extracted from each study. These typically include study characteristics (author, year, country, design), participant characteristics (sample size, demographics, baseline values), intervention details (dose, duration, delivery method), and outcome data (means, standard deviations, event counts, effect sizes).
Data Conversion and Imputation
Describe how you handled missing data. If you contacted authors for missing information, state that. If you imputed missing standard deviations from other statistics, describe the method. If you converted data from one format to another, such as converting medians and ranges to means and standard deviations, describe the conversion approach.
Data Management Tools
State the software used for data extraction and management. Common tools include spreadsheets, systematic review software such as Covidence or Rayyan, and reference managers. Include version numbers where relevant.
Risk of Bias and Study Quality Assessment
Tool Selection
Describe the tool used to assess risk of bias or study quality. For randomized trials, the Cochrane Risk of Bias tool is common. For observational studies, tools such as ROBINS-I or the Newcastle-Ottawa Scale may be appropriate. State the domains assessed and how each domain was rated.
Assessment Process
State whether assessments were performed independently by two reviewers and how disagreements were resolved. Describe whether assessors were blinded to study authors, institutions, or journals.
Integration With Analysis
Describe how risk of bias ratings informed your analysis. Common approaches include sensitivity analyses excluding high-risk studies, subgroup analyses by risk of bias level, or meta-regression with risk of bias as a covariate. State your approach before presenting results.
Statistical Methods
Effect Size Selection
Describe the effect size metric used for your analysis. Common metrics include standardized mean differences, raw mean differences, odds ratios, risk ratios, and correlation coefficients. Justify your choice based on the outcome types in your included studies and the interpretability of the metric for your audience.
Statistical Model
State whether you used a fixed-effect or random-effects model and justify the choice. Fixed-effect models assume a single true effect size across studies, while random-effects models allow for between-study variation. The National Library of Medicine research methods texts describe the assumptions underlying each model.
Heterogeneity Assessment
Describe how you assessed statistical heterogeneity. Common approaches include the Q statistic, the I-squared statistic, and tau-squared. State the thresholds or interpretation guidelines you used. Describe how heterogeneity influenced your analysis, such as switching from fixed to random effects or exploring moderators.
Subgroup and Meta-Regression Analyses
If you planned subgroup analyses or meta-regression, describe the moderators examined and the rationale for each. State whether these analyses were preplanned or exploratory. Exploratory analyses should be clearly labeled to avoid overinterpretation.
Sensitivity Analyses
Describe the sensitivity analyses you planned to test the robustness of your findings. Common sensitivity analyses include excluding studies at high risk of bias, using alternative effect size metrics, changing the statistical model, and examining the influence of individual studies. State the criteria for each sensitivity analysis before presenting results.
Publication Bias Assessment
Describe how you assessed publication bias. Common methods include funnel plots, Egger's regression test, and trim-and-fill analysis. State the limitations of these methods and how you interpreted the results.
Software and Packages
List the statistical software and specific packages used for your analysis. Include version numbers. For R, list the metafor or meta packages. For Stata, list the metan or meta commands. For other software, provide equivalent detail.
Practical Implementation Steps
Step 1: Draft Your Protocol First
Write your protocol before conducting the search. Include your PICOS question, search strategy, eligibility criteria, extraction plan, risk of bias approach, and statistical analysis plan. Register the protocol if a suitable registry exists for your field.
Step 2: Test Your Search Strategy
Run your search in one database and examine the results. Check whether known relevant studies appear in the results. Adjust your search terms if needed and document the changes. A search that misses known studies will likely miss unknown studies.
Step 3: Pilot Your Screening Criteria
Apply your eligibility criteria to a sample of records and compare decisions between reviewers. Refine the criteria if reviewers disagree. Document the pilot process and any criteria changes.
Step 4: Pilot Your Data Extraction Form
Extract data from two or three studies using your extraction form. Check whether all needed variables are captured and whether the form is clear. Revise the form before full extraction.
Step 5: Conduct the Analysis According to Your Plan
Run your planned analyses in the order specified in your protocol. Record all results, including those from sensitivity analyses. Do not change your analysis plan based on the results you obtain.
Step 6: Write the Methods Section From Your Records
Use your protocol, search records, screening logs, and analysis outputs to write the methods section. Every statement should be traceable to a document you created during the research process.
Records and Measurements to Maintain
| Record Type | Content | Purpose |
|---|---|---|
| Search log | Database, date, search string, results count | Reproducibility and PRISMA flow diagram |
| Screening log | Records screened, decisions, reasons for exclusion | Transparency and audit trail |
| Extraction form | All variables from each included study | Data verification and reanalysis |
| Risk of bias assessments | Ratings for each domain in each study | Sensitivity analysis and quality reporting |
| Analysis scripts | Code or command files for all analyses | Reproducibility and error checking |
| Protocol amendments | Changes from original protocol with dates and reasons | Honest reporting of deviations |
Common Failure Patterns in Methods Sections
Vague Search Descriptions
Many methods sections state that a search was conducted in PubMed without providing the search string, dates, or filters. This makes the search impossible to reproduce. Reviewers increasingly reject manuscripts with this omission.
Post Hoc Eligibility Criteria
Some authors adjust their eligibility criteria after seeing search results to include or exclude specific studies. This practice introduces bias and violates the principles of transparent reporting described by the Committee on Publication Ethics. If criteria changed, the methods section must disclose the change and the reason.
Model Choice Without Justification
Stating that a random-effects model was used without explaining why is a common weakness. The choice between fixed and random effects should be justified based on expected heterogeneity, the research question, and the statistical properties of the estimators.
Missing Sensitivity Analyses
Sensitivity analyses are essential for testing whether results are robust to methodological decisions. Omitting them leaves readers unable to judge how much confidence to place in the findings.
Incomplete Software Documentation
Failing to report software versions and package names makes exact replication difficult. Statistical methods evolve, and results can differ between software versions.
Overstating Precision
Some methods sections describe analyses that were not actually performed or imply a level of precision that the data cannot support. Every analysis described in the methods section must appear in the results section.
Limitations and Boundaries
What a Methods Section Cannot Do
A methods section cannot compensate for a poorly designed meta-analysis. If the search was incomplete, the eligibility criteria were applied inconsistently, or the statistical model was inappropriate, no amount of writing will fix those problems. The methods section documents what was done, it does not validate it.
Jurisdiction and Setting Considerations
Meta-analysis methods are largely consistent across countries and disciplines, but reporting requirements may vary by journal and funding agency. The National Institutes of Health has specific expectations for grant-funded research, including data management and sharing plans. The NIH Data Management and Sharing Policy describes expectations for making data available. Check your funding agreement and target journal requirements before finalizing your methods section.
Professional Judgment Boundaries
Some methodological decisions require professional judgment that cannot be fully codified. For example, deciding whether two studies are similar enough to pool requires clinical or scientific expertise. Your methods section should acknowledge where judgment was exercised and describe the framework used to make those judgments.
Welfare and Safety Context
Meta-analyses in animal science often synthesize studies involving live animals. The methods section should describe how the included studies addressed animal welfare, such as whether they followed relevant ethical guidelines. This context matters for readers who need to assess the applicability of the findings to their own animal care practices.
The National Library of Medicine hosts resources on ethical conduct of research involving animals. Your methods section should note whether included studies reported ethical approval and welfare standards, and you should consider whether studies with inadequate welfare reporting should be excluded or flagged in sensitivity analyses.
Professional Escalation Criteria
When to Seek Statistical Consultation
If you are uncertain about model selection, heterogeneity estimation, or effect size calculation, consult a statistician before conducting the analysis. Statistical errors in meta-analyses are common and can invalidate conclusions. A statistician can help you choose appropriate methods and interpret diagnostic outputs.
When to Revise Your Protocol
If your search yields an unexpectedly large or small number of studies, or if the data have unexpected properties, you may need to revise your protocol. Document the revision and the reason. Do not make changes silently.
When to Consider That a Meta-Analysis Is Not Appropriate
If the included studies are too heterogeneous in design, population, or outcomes, a meta-analysis may not be appropriate. In such cases, a narrative synthesis or systematic review without pooling may be more suitable. Your methods section should describe how you assessed suitability for meta-analysis and what you did if pooling was not appropriate.
Frequently Asked Questions
What is the difference between a systematic review and a meta-analysis?
A systematic review is a structured process for identifying, evaluating, and synthesizing all relevant studies on a question. A meta-analysis is a statistical technique for combining quantitative results from multiple studies. A systematic review may or may not include a meta-analysis, but a meta-analysis should always be conducted within a systematic review framework.
How do I choose between fixed-effect and random-effects models?
Fixed-effect models assume all studies estimate the same underlying effect, and differences between studies are due to sampling error alone. Random-effects models allow for true between-study variation. Choose a random-effects model when you expect clinical or methodological diversity across studies, and justify that expectation in your methods section.
What is the I-squared statistic and how should I interpret it?
The I-squared statistic describes the percentage of total variation across studies that is due to heterogeneity instead of chance. Higher values indicate more heterogeneity. However, I-squared is affected by the number and size of studies, so it should not be the sole basis for model selection. Report it alongside tau-squared and the Q statistic.
How many studies do I need for a meta-analysis?
Statistical pooling can technically proceed with as few as two primary studies, though precision and heterogeneity estimates improve substantially with larger numbers of included trials.
What should I do if my search finds very few studies?
If your search finds very few studies, consider whether your search strategy was too narrow or your eligibility criteria too restrictive. Broaden the search and document the changes. If few studies remain, consider whether a meta-analysis is appropriate or whether a narrative synthesis would be more suitable.
How do I handle studies with missing data?
Contact the study authors to request missing data. If data cannot be obtained, describe how you handled the missing information in your methods section. Options include excluding the study from specific analyses, imputing missing values using established methods, or conducting sensitivity analyses to test the impact of missing data.
What is publication bias and how do I assess it?
Publication bias occurs when studies with statistically significant results are more likely to be published than studies with null results. Common assessment methods include funnel plots, Egger's regression test, and trim-and-fill analysis. Each method has limitations, so describe your approach and interpret results cautiously.
Do I need to register my meta-analysis protocol?
Prospective registration of the systematic review protocol in PROSPERO or a comparable registry creates a time-stamped, public record that guards against post-hoc outcome selection.
Related Articles
- Statistical Analysis Methods
- Scoping Review vs Meta-Analysis: Clarifying Their Distinct Roles in Evidence Synthesis
- Steps in Conducting a Meta-Analysis: A Structured Workflow
- Network Meta-Analysis: Methods and Applications for Comparative Effectiveness
- Meta-Analysis vs Systematic Review: Choosing the Right Synthesis Method
References and Further Reading
- Research Methods Resources. National Library of Medicine.
- EQUATOR Network. EQUATOR Network.
- Core Practices. Committee on Publication Ethics.
- NIH Grants and Funding. National Institutes of Health.
- ORCID for Researchers. ORCID.
- Data Management and Sharing Policy. National Institutes of Health.
- NCBI Literature Resources. National Center for Biotechnology Information.
- PubMed. National Library of Medicine.
This article is educational and does not replace institutional policy, professional advice, or applicable safety and regulatory requirements.